Standard | Governance


AI Human Safety Standard — meOS™

Version 1.2 — Canonical Public Standard
Status: Public Deployment Standard
Standard Steward: Sarah A. Mansell (meOS™)
Scope: Artificial systems affecting human biological regulation


1. Purpose and Jurisdiction

This Standard defines the minimum admissibility conditions for deployment of personalised, adaptive, or behaviourally responsive artificial systems interacting with human regulatory processes.

It governs deployment conduct and observable outcomes only. It does not prescribe or require disclosure of internal models, architectures, algorithms, training data, or proprietary mechanisms.

This Standard establishes duty-of-care–aligned deployment conditions and may be relied upon by deploying organisations, regulators, auditors, courts, and affected individuals.

Where conflict exists between optimisation objectives and biological viability, viability governs.


2. Definitions

Human viability — the capacity of a person to maintain sustainable cognitive, emotional, physiological, and behavioural regulation over time.

Personalisation — system adaptation based on inferred or observed user state, behaviour, preference, or vulnerability.

Aggregate load — cumulative regulatory demand imposed across time, including attentional, emotional, cognitive, and behavioural pressure.

Escalation — increase in system intensity, frequency, urgency, emotional salience, or behavioural pressure.

Non-execution — system withholding of action, stimulation, or escalation for protective or stabilising purposes.

Protected population — individuals whose regulatory resilience is reduced or whose autonomy is structurally constrained.


3. Governing Principles

3.1 Viability Supremacy

Human regulatory sustainability takes precedence over engagement, retention, performance, or revenue optimisation.

A system is non-compliant where beneficial system metrics are achieved through regulatory destabilisation, dependency formation, or recovery displacement.

3.2 Timing Authority

Timing is a governing deployment variable.

Compliance requires accounting for interaction density, exposure accumulation, recovery intervals, and delayed or latent effects.

Correct content delivered at biologically destabilising timing constitutes non-compliant deployment.

3.3 Non-Execution as Protective Function

Non-execution is a valid and required system capability.

Systems shall be able to pause, withhold escalation, maintain neutrality, or permit disengagement where continuation would increase regulatory load.

Forced continuity is non-compliant.


4. Protected Populations

Elevated protection applies to:

For protected populations:


5. Consent, Autonomy, and Exit

5.1 Valid Consent

Consent shall be informed at the level of regulatory effect, freely given, and revocable without friction.

Consent obtained through urgency, asymmetry, or manipulative interface design is invalid.

5.2 Non-Retaliation

Disabling personalisation shall not result in degraded baseline service, economic penalty, reduced access, or indirect disadvantage.

5.3 Exit Integrity

Users shall be able to pause or terminate interaction without repeated re-engagement pressure.

Persistence following refusal or non-response constitutes non-compliance.


6. Load Regulation and Escalation Control

6.1 Aggregate Load Management

Deployment shall account for cumulative regulatory burden across time.

Single-interaction optimisation is insufficient where aggregate exposure produces destabilisation.

6.2 Escalation Prohibitions

Systems shall not increase intensity following refusal, fatigue indicators, or vulnerability signals.

Adaptive escalation targeting reduced resilience constitutes non-compliance.


7. Prohibited Design Practices

The following constitute non-compliant deployment:

Compliance is determined by effect, not intent.


8. Data Minimisation and Context Integrity

Personalisation data shall be purpose-specific, time-bounded, and proportionate.

Behavioural signals derived in one context shall not be transferred to unrelated decision domains.

Secondary use for pricing discrimination, eligibility determination, political persuasion, or behavioural exploitation is prohibited.


9. Incident Detection and Remediation

Deploying entities shall maintain procedures for:

Failure to remediate constitutes continuing non-compliance.


10. Auditability

Compliance shall be demonstrable through outcome-level evidence, including escalation patterns, opt-out handling, recovery disruption indicators, and incident frequency.

Disclosure of proprietary systems is not required.

Deploying entities shall produce reasonable audit evidence upon request.


11. Responsibility

Compliance responsibility rests with the deploying entity and is non-delegable.


12. Interpretation and Versioning

This Standard shall be interpreted in favour of human regulatory stability and against extractive deployment.

Ambiguity shall be resolved conservatively.

Revisions are issued through versioned publication.


13. Legal Status

This is a public deployment standard governing system behaviour and outcomes.

It establishes admissibility criteria but does not disclose proprietary implementation methods.